A project based on: Growing Neural Cellular Automata
Realized during the course Physically-Based Simulation in Computer Graphics HS2022 @ETHZ.
All the .ipynb files are also avaible on google collab. To access check the links below:
- mainNCA.ipynb: Most of the code is written here, it includes fluid and collision animations. The dataset is simulated locally and exported to folders in
data/exportwhich the notebook then accesses remotely. - dlca_with_sim.ipynb: This version branched off from the main as it uses another approach where the simulation code is integrated into the code to accelerate the runtime. This version achieved the best results to predict simulation steps for incompressible fluids.
- scaling for 256: In this notebook we show our experiments in using larger grids for the NCA fluid simulation.
- styling NCA: This Notebook use the styles NCA to update the fluid NCA channels to give to the fluid some targetted style
If you decide to run the notebooks the a dataset generation script locally, make sure to remove (comment out) all commands starting with an ! from the notebook and install the required modules:
- tensorflow
- numpy
- taichi
- mediapy
- matplotlib
- PIL
- tqdm
to install ffmpeg use the following command (ubuntu):
command -v ffmpeg >/dev/null || (apt update && apt install -y ffmpeg)
The final presentation can be viewed here or be accessed via figma, which is recommended as the GIF animations are not shown in the PDF file:
Salimbeni Etienne Alain Jaroslav, Andrin Rehmann